choice-metrics.com

Dear colleagues

I have run an unlabelled choice experiment with 2 alternatives + SQ option, 2 attributes A and B + 1 monetary attribute C.
Attribute A has 3 levels, dummy-coded 0, 1, 2
Attribute B has 2 levels:, dummy-coded 0, 1
Attribute C has 5 levels
I asked respondents what are their current practices with respect to A and B to caraterize individual-specific SQ.
My problem is the following. For some respondents, the SQ is different from the levels of attributes A and B. How should I code their SQ? Can I use -1 to keep 0 as the reference level ?

For more context, attributes A and B related to the nature of the data their are sharing with a manager, and the levels depends on the richness of these data and whether it's transmitted automatically or not. 0 corresponds to manual, basic data sharing for both attributes A and B. However, for a minority of respondents, the SQ is no data sharing.

Thanks for your help
Marianne

It sounds like you need an additional variable "datasharing" that has levels 0 (no sharing) or 1 (sharing).
Then you would have the following variables in your utility function: datasharing, A, B, C

Your utility function could then look something like the following (using syntax in Apollo or Biogeme):

Code: Select all

U = b1   * (datasharing==1)
  + b2_1 * (datasharing==1) * (A==1) + b2_2 * (datasharing==1) * (A==2)
  + b3_1 * (datasharing==1) * (B==1)
  + b4   * C

In other words, these interactions result in the following:
if datasharing==0, then U=0 + b4*C,
if datasharing==1 and A==2 and B==1, then U=b1+b2_2 + b3_1 + b4*C

Michiel